Dispatch CLI

Dispatch Agents is a platform for building and deploying AI agents, currently in public preview at dispatchagents.ai. This CLI manages the full agent lifecycle — from scaffolding and local development to containerization and cloud deployment.

Installation

From GitHub (Recommended)

# Get the latest CLI version tag
git ls-remote --tags git@github.com:datadog-labs/dispatch_agents_cli.git 'v*' | sort -t'/' -k3 -V | tail -1 | awk -F'/' '{print $3}'

# Install using the version tag (replace vX.Y.Z with the output above)
uv tool install git+ssh://git@github.com/datadog-labs/dispatch_agents_cli.git@vX.Y.Z --upgrade

Tip: After installing, run dispatch update-cli to check for newer versions.

Quick Start

# Initialize and run a single agent
dispatch agent init
dispatch agent run                                          # auto-builds if needed
dispatch agent send-event --topic "test" --payload "Hello"

# Multi-agent setup
cd my-first-agent/ && dispatch agent init && dispatch agent register --topics "data-analysis"
cd my-second-agent/ && dispatch agent init && dispatch agent register --topics "chat"
dispatch router start
dispatch router test --topic "chat" --payload "Hello world"
dispatch router stop

Configuration

Configuration is stored in dispatch.yaml (generated by dispatch agent init). All configuration options:

# Required fields
namespace: my-namespace                 # Deployment namespace (contact your org admin)
agent_name: my-agent                   # Agent identifier
entrypoint: agent.py                   # Python file with @on/@fn decorated handlers
base_image: python:3.13-slim           # Docker base image

# Optional fields
system_packages:                       # Additional system packages to install (apt)
  - curl

local_dependencies:                    # Local packages for monorepo development (name -> path)
  my_package: ../path/to/package

env:                                   # Plain environment variables; values must be strings
  LOG_LEVEL: "debug"

secrets:                               # Secrets injected from secrets manager as env vars
  - name: DD_API_KEY
    secret_id: /datadog-agent/dd-api-key

volumes:                               # Persistent storage volumes (survive container restarts)
  - name: data
    mountPath: /data
    mode: read_write_many

mcp_servers:                           # MCP servers from the platform registry
  - server: datadog

resources:                             # Container resource limits
  limits:
    cpu: "500m"                        # e.g. "250m", "500m", "1", "2"
    memory: "1Gi"                      # e.g. "512Mi", "1Gi", "2Gi"

env: Platform-reserved names (BACKEND_URL, DISPATCH_API_KEY, etc.) cannot be used as keys. secrets: LLM provider keys (OPENAI_API_KEY, ANTHROPIC_API_KEY, etc.) are managed by the LLM gateway — no need to list them here. resources: Defaults are 250m CPU and 2Gi memory. Values must form a valid ECS Fargate combination.

Commands

Run dispatch --help or dispatch <command> --help for the full command reference.

Skills Hub

Skills are reusable SKILL.md files that provide instructions for AI agents, designed for sharing domain expertise and standardized behaviors across teams.

The fastest way to get started is with the Claude Code plugin:

claude plugin marketplace add datadog-labs/dispatch_agents_sdk && claude plugin install dispatch-agents@dispatch-agents
/dispatch-agents:getting-started

All dispatch skills commands accept --namespace to target a specific namespace. Without it, the namespace resolves from the --namespace flag, DISPATCH_NAMESPACE env var, or dispatch.yaml in the current directory.

See SKILL.md for the file format specification.

Support

• GitHub Issues: github.com/datadog-labs/dispatch_agents_cli/issues